2 * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "dpif-linux.h"
27 #include <linux/types.h>
28 #include <linux/pkt_sched.h>
29 #include <linux/rtnetlink.h>
30 #include <linux/sockios.h>
36 #include "dpif-provider.h"
37 #include "dynamic-string.h"
40 #include "netdev-linux.h"
41 #include "netdev-vport.h"
42 #include "netlink-socket.h"
46 #include "openvswitch/tunnel.h"
48 #include "poll-loop.h"
49 #include "rtnetlink.h"
50 #include "rtnetlink-link.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(dpif_linux);
59 enum { LRU_MAX_PORTS = 1024 };
60 enum { LRU_MASK = LRU_MAX_PORTS - 1};
61 BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
63 struct dpif_linux_dp {
64 /* Generic Netlink header. */
67 /* struct odp_header. */
71 const char *name; /* ODP_DP_ATTR_NAME. */
72 struct odp_stats stats; /* ODP_DP_ATTR_STATS. */
73 enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */
74 const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */
75 uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */
78 static void dpif_linux_dp_init(struct dpif_linux_dp *);
79 static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
80 const struct ofpbuf *);
81 static void dpif_linux_dp_dump_start(struct nl_dump *);
82 static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
83 struct dpif_linux_dp *reply,
84 struct ofpbuf **bufp);
85 static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
86 struct ofpbuf **bufp);
88 struct dpif_linux_flow {
89 /* Generic Netlink header. */
92 /* struct odp_header. */
93 unsigned int nlmsg_flags;
98 * The 'stats' and 'used' members point to 64-bit data that might only be
99 * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to
100 * access their values.
102 * If 'actions' is nonnull then ODP_FLOW_ATTR_ACTIONS will be included in
103 * the Netlink version of the command, even if actions_len is zero. */
104 const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */
106 const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */
108 const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */
109 const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */
110 const uint64_t *used; /* ODP_FLOW_ATTR_USED. */
111 bool clear; /* ODP_FLOW_ATTR_CLEAR. */
114 static void dpif_linux_flow_init(struct dpif_linux_flow *);
115 static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *,
116 const struct ofpbuf *);
117 static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
119 static int dpif_linux_flow_transact(const struct dpif_linux_flow *request,
120 struct dpif_linux_flow *reply,
121 struct ofpbuf **bufp);
122 static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
123 struct dpif_flow_stats *);
125 /* Datapath interface for the openvswitch Linux kernel module. */
130 /* Multicast group messages. */
131 struct nl_sock *mc_sock;
132 uint32_t mcgroups[DPIF_N_UC_TYPES];
133 unsigned int listen_mask;
135 /* Change notification. */
136 struct sset changed_ports; /* Ports that have changed. */
137 struct rtnetlink_notifier port_notifier;
140 /* Queue of unused ports. */
141 unsigned long *lru_bitmap;
142 uint16_t lru_ports[LRU_MAX_PORTS];
147 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
149 /* Generic Netlink family numbers for ODP. */
150 static int odp_datapath_family;
151 static int odp_vport_family;
152 static int odp_flow_family;
153 static int odp_packet_family;
155 /* Generic Netlink socket. */
156 static struct nl_sock *genl_sock;
158 static int dpif_linux_init(void);
159 static int open_dpif(const struct dpif_linux_dp *, struct dpif **);
160 static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
163 static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
165 static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *,
166 const struct ofpbuf *);
168 static struct dpif_linux *
169 dpif_linux_cast(const struct dpif *dpif)
171 dpif_assert_class(dpif, &dpif_linux_class);
172 return CONTAINER_OF(dpif, struct dpif_linux, dpif);
176 dpif_linux_push_port(struct dpif_linux *dp, uint16_t port)
178 if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) {
179 bitmap_set1(dp->lru_bitmap, port);
180 dp->lru_ports[dp->lru_head++ & LRU_MASK] = port;
185 dpif_linux_pop_port(struct dpif_linux *dp)
189 if (dp->lru_head == dp->lru_tail) {
193 port = dp->lru_ports[dp->lru_tail++ & LRU_MASK];
194 bitmap_set0(dp->lru_bitmap, port);
199 dpif_linux_enumerate(struct sset *all_dps)
205 error = dpif_linux_init();
210 dpif_linux_dp_dump_start(&dump);
211 while (nl_dump_next(&dump, &msg)) {
212 struct dpif_linux_dp dp;
214 if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
215 sset_add(all_dps, dp.name);
218 return nl_dump_done(&dump);
222 dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
223 bool create, struct dpif **dpifp)
225 struct dpif_linux_dp dp_request, dp;
229 error = dpif_linux_init();
234 /* Create or look up datapath. */
235 dpif_linux_dp_init(&dp_request);
236 dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET;
237 dp_request.name = name;
238 error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
242 error = open_dpif(&dp, dpifp);
249 open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
251 struct dpif_linux *dpif;
255 dpif = xmalloc(sizeof *dpif);
256 error = rtnetlink_link_notifier_register(&dpif->port_notifier,
257 dpif_linux_port_changed, dpif);
262 dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
263 dp->dp_ifindex, dp->dp_ifindex);
265 dpif->mc_sock = NULL;
266 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
267 dpif->mcgroups[i] = dp->mcgroups[i];
269 dpif->listen_mask = 0;
270 dpif->dp_ifindex = dp->dp_ifindex;
271 sset_init(&dpif->changed_ports);
272 dpif->change_error = false;
273 *dpifp = &dpif->dpif;
275 dpif->lru_head = dpif->lru_tail = 0;
276 dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
277 bitmap_set1(dpif->lru_bitmap, ODPP_LOCAL);
278 for (i = 1; i < LRU_MAX_PORTS; i++) {
279 dpif_linux_push_port(dpif, i);
289 dpif_linux_close(struct dpif *dpif_)
291 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
292 rtnetlink_link_notifier_unregister(&dpif->port_notifier);
293 sset_destroy(&dpif->changed_ports);
294 free(dpif->lru_bitmap);
299 dpif_linux_destroy(struct dpif *dpif_)
301 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
302 struct dpif_linux_dp dp;
304 dpif_linux_dp_init(&dp);
305 dp.cmd = ODP_DP_CMD_DEL;
306 dp.dp_ifindex = dpif->dp_ifindex;
307 return dpif_linux_dp_transact(&dp, NULL, NULL);
311 dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
313 struct dpif_linux_dp dp;
317 error = dpif_linux_dp_get(dpif_, &dp, &buf);
326 dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
328 struct dpif_linux_dp dp;
332 error = dpif_linux_dp_get(dpif_, &dp, &buf);
334 *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP;
341 dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
343 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
344 struct dpif_linux_dp dp;
346 dpif_linux_dp_init(&dp);
347 dp.cmd = ODP_DP_CMD_SET;
348 dp.dp_ifindex = dpif->dp_ifindex;
349 dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO;
350 return dpif_linux_dp_transact(&dp, NULL, NULL);
354 dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
357 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
358 const char *name = netdev_get_name(netdev);
359 const char *type = netdev_get_type(netdev);
360 struct dpif_linux_vport request, reply;
361 const struct ofpbuf *options;
365 dpif_linux_vport_init(&request);
366 request.cmd = ODP_VPORT_CMD_NEW;
367 request.dp_ifindex = dpif->dp_ifindex;
368 request.type = netdev_vport_get_vport_type(netdev);
369 if (request.type == ODP_VPORT_TYPE_UNSPEC) {
370 VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has "
371 "unsupported type `%s'",
372 dpif_name(dpif_), name, type);
377 options = netdev_vport_get_options(netdev);
378 if (options && options->size) {
379 request.options = options->data;
380 request.options_len = options->size;
383 /* Loop until we find a port that isn't used. */
385 request.port_no = dpif_linux_pop_port(dpif);
386 error = dpif_linux_vport_transact(&request, &reply, &buf);
389 *port_nop = reply.port_no;
392 } while (request.port_no != UINT32_MAX
393 && (error == EBUSY || error == EFBIG));
399 dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
401 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
402 struct dpif_linux_vport vport;
405 dpif_linux_vport_init(&vport);
406 vport.cmd = ODP_VPORT_CMD_DEL;
407 vport.dp_ifindex = dpif->dp_ifindex;
408 vport.port_no = port_no;
409 error = dpif_linux_vport_transact(&vport, NULL, NULL);
412 dpif_linux_push_port(dpif, port_no);
418 dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no,
419 const char *port_name, struct dpif_port *dpif_port)
421 struct dpif_linux_vport request;
422 struct dpif_linux_vport reply;
426 dpif_linux_vport_init(&request);
427 request.cmd = ODP_VPORT_CMD_GET;
428 request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex;
429 request.port_no = port_no;
430 request.name = port_name;
432 error = dpif_linux_vport_transact(&request, &reply, &buf);
434 dpif_port->name = xstrdup(reply.name);
435 dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
436 dpif_port->port_no = reply.port_no;
438 netdev_stats_from_rtnl_link_stats64(&dpif_port->stats,
441 memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats);
449 dpif_linux_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
450 struct dpif_port *dpif_port)
452 return dpif_linux_port_query__(dpif, port_no, NULL, dpif_port);
456 dpif_linux_port_query_by_name(const struct dpif *dpif, const char *devname,
457 struct dpif_port *dpif_port)
459 return dpif_linux_port_query__(dpif, 0, devname, dpif_port);
463 dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
465 /* If the datapath increases its range of supported ports, then it should
466 * start reporting that. */
471 dpif_linux_flow_flush(struct dpif *dpif_)
473 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
474 struct dpif_linux_flow flow;
476 dpif_linux_flow_init(&flow);
477 flow.cmd = ODP_FLOW_CMD_DEL;
478 flow.dp_ifindex = dpif->dp_ifindex;
479 return dpif_linux_flow_transact(&flow, NULL, NULL);
482 struct dpif_linux_port_state {
484 unsigned long *port_bitmap; /* Ports in the datapath. */
485 bool complete; /* Dump completed without error. */
489 dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep)
491 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
492 struct dpif_linux_port_state *state;
493 struct dpif_linux_vport request;
496 *statep = state = xmalloc(sizeof *state);
497 state->port_bitmap = bitmap_allocate(LRU_MAX_PORTS);
498 state->complete = false;
500 dpif_linux_vport_init(&request);
501 request.cmd = ODP_DP_CMD_GET;
502 request.dp_ifindex = dpif->dp_ifindex;
504 buf = ofpbuf_new(1024);
505 dpif_linux_vport_to_ofpbuf(&request, buf);
506 nl_dump_start(&state->dump, genl_sock, buf);
513 dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_,
514 struct dpif_port *dpif_port)
516 struct dpif_linux_port_state *state = state_;
517 struct dpif_linux_vport vport;
521 if (!nl_dump_next(&state->dump, &buf)) {
522 state->complete = true;
526 error = dpif_linux_vport_from_ofpbuf(&vport, &buf);
531 if (vport.port_no < LRU_MAX_PORTS) {
532 bitmap_set1(state->port_bitmap, vport.port_no);
535 dpif_port->name = (char *) vport.name;
536 dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
537 dpif_port->port_no = vport.port_no;
539 netdev_stats_from_rtnl_link_stats64(&dpif_port->stats, vport.stats);
541 memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats);
547 dpif_linux_port_dump_done(const struct dpif *dpif_, void *state_)
549 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
550 struct dpif_linux_port_state *state = state_;
551 int error = nl_dump_done(&state->dump);
553 if (state->complete) {
556 for (i = 0; i < LRU_MAX_PORTS; i++) {
557 if (!bitmap_is_set(state->port_bitmap, i)) {
558 dpif_linux_push_port(dpif, i);
563 free(state->port_bitmap);
569 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
571 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
573 if (dpif->change_error) {
574 dpif->change_error = false;
575 sset_clear(&dpif->changed_ports);
577 } else if (!sset_is_empty(&dpif->changed_ports)) {
578 *devnamep = sset_pop(&dpif->changed_ports);
586 dpif_linux_port_poll_wait(const struct dpif *dpif_)
588 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
589 if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) {
590 poll_immediate_wake();
592 rtnetlink_link_notifier_wait();
597 dpif_linux_flow_get__(const struct dpif *dpif_,
598 const struct nlattr *key, size_t key_len,
599 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
601 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
602 struct dpif_linux_flow request;
604 dpif_linux_flow_init(&request);
605 request.cmd = ODP_FLOW_CMD_GET;
606 request.dp_ifindex = dpif->dp_ifindex;
608 request.key_len = key_len;
609 return dpif_linux_flow_transact(&request, reply, bufp);
613 dpif_linux_flow_get(const struct dpif *dpif_,
614 const struct nlattr *key, size_t key_len,
615 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
617 struct dpif_linux_flow reply;
621 error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf);
624 dpif_linux_flow_get_stats(&reply, stats);
627 buf->data = (void *) reply.actions;
628 buf->size = reply.actions_len;
638 dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
639 const struct nlattr *key, size_t key_len,
640 const struct nlattr *actions, size_t actions_len,
641 struct dpif_flow_stats *stats)
643 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
644 struct dpif_linux_flow request, reply;
645 struct nlattr dummy_action;
649 dpif_linux_flow_init(&request);
650 request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET;
651 request.dp_ifindex = dpif->dp_ifindex;
653 request.key_len = key_len;
654 /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */
655 request.actions = actions ? actions : &dummy_action;
656 request.actions_len = actions_len;
657 if (flags & DPIF_FP_ZERO_STATS) {
658 request.clear = true;
660 request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
661 error = dpif_linux_flow_transact(&request,
662 stats ? &reply : NULL,
663 stats ? &buf : NULL);
664 if (!error && stats) {
665 dpif_linux_flow_get_stats(&reply, stats);
672 dpif_linux_flow_del(struct dpif *dpif_,
673 const struct nlattr *key, size_t key_len,
674 struct dpif_flow_stats *stats)
676 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
677 struct dpif_linux_flow request, reply;
681 dpif_linux_flow_init(&request);
682 request.cmd = ODP_FLOW_CMD_DEL;
683 request.dp_ifindex = dpif->dp_ifindex;
685 request.key_len = key_len;
686 error = dpif_linux_flow_transact(&request,
687 stats ? &reply : NULL,
688 stats ? &buf : NULL);
689 if (!error && stats) {
690 dpif_linux_flow_get_stats(&reply, stats);
696 struct dpif_linux_flow_state {
698 struct dpif_linux_flow flow;
699 struct dpif_flow_stats stats;
704 dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep)
706 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
707 struct dpif_linux_flow_state *state;
708 struct dpif_linux_flow request;
711 *statep = state = xmalloc(sizeof *state);
713 dpif_linux_flow_init(&request);
714 request.cmd = ODP_DP_CMD_GET;
715 request.dp_ifindex = dpif->dp_ifindex;
717 buf = ofpbuf_new(1024);
718 dpif_linux_flow_to_ofpbuf(&request, buf);
719 nl_dump_start(&state->dump, genl_sock, buf);
728 dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_,
729 const struct nlattr **key, size_t *key_len,
730 const struct nlattr **actions, size_t *actions_len,
731 const struct dpif_flow_stats **stats)
733 struct dpif_linux_flow_state *state = state_;
738 ofpbuf_delete(state->buf);
741 if (!nl_dump_next(&state->dump, &buf)) {
745 error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf);
750 if (actions && !state->flow.actions) {
751 error = dpif_linux_flow_get__(dpif_, state->flow.key,
753 &state->flow, &state->buf);
754 if (error == ENOENT) {
755 VLOG_DBG("dumped flow disappeared on get");
757 VLOG_WARN("error fetching dumped flow: %s", strerror(error));
763 *actions = state->flow.actions;
764 *actions_len = state->flow.actions_len;
767 *key = state->flow.key;
768 *key_len = state->flow.key_len;
771 dpif_linux_flow_get_stats(&state->flow, &state->stats);
772 *stats = &state->stats;
778 dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
780 struct dpif_linux_flow_state *state = state_;
781 int error = nl_dump_done(&state->dump);
782 ofpbuf_delete(state->buf);
788 dpif_linux_execute__(int dp_ifindex,
789 const struct nlattr *key, size_t key_len,
790 const struct nlattr *actions, size_t actions_len,
791 const struct ofpbuf *packet)
793 struct odp_header *execute;
797 buf = ofpbuf_new(128 + actions_len + packet->size);
799 nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
800 ODP_PACKET_CMD_EXECUTE, 1);
802 execute = ofpbuf_put_uninit(buf, sizeof *execute);
803 execute->dp_ifindex = dp_ifindex;
805 nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size);
806 nl_msg_put_unspec(buf, ODP_PACKET_ATTR_KEY, key, key_len);
807 nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len);
809 error = nl_sock_transact(genl_sock, buf, NULL);
815 dpif_linux_execute(struct dpif *dpif_,
816 const struct nlattr *key, size_t key_len,
817 const struct nlattr *actions, size_t actions_len,
818 const struct ofpbuf *packet)
820 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
822 return dpif_linux_execute__(dpif->dp_ifindex, key, key_len,
823 actions, actions_len, packet);
827 dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
829 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
830 *listen_mask = dpif->listen_mask;
835 dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
837 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
841 if (listen_mask == dpif->listen_mask) {
843 } else if (!listen_mask) {
844 nl_sock_destroy(dpif->mc_sock);
845 dpif->mc_sock = NULL;
846 dpif->listen_mask = 0;
848 } else if (!dpif->mc_sock) {
849 error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock);
855 /* Unsubscribe from old groups. */
856 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
857 if (dpif->listen_mask & (1u << i)) {
858 nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
862 /* Update listen_mask. */
863 dpif->listen_mask = listen_mask;
865 /* Subscribe to new groups. */
867 for (i = 0; i < DPIF_N_UC_TYPES; i++) {
868 if (dpif->listen_mask & (1u << i)) {
871 retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
881 dpif_linux_get_sflow_probability(const struct dpif *dpif_,
882 uint32_t *probability)
884 struct dpif_linux_dp dp;
888 error = dpif_linux_dp_get(dpif_, &dp, &buf);
890 *probability = dp.sampling ? *dp.sampling : 0;
897 dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
899 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
900 struct dpif_linux_dp dp;
902 dpif_linux_dp_init(&dp);
903 dp.cmd = ODP_DP_CMD_SET;
904 dp.dp_ifindex = dpif->dp_ifindex;
905 dp.sampling = &probability;
906 return dpif_linux_dp_transact(&dp, NULL, NULL);
910 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
911 uint32_t queue_id, uint32_t *priority)
913 if (queue_id < 0xf000) {
914 *priority = TC_H_MAKE(1 << 16, queue_id + 1);
922 parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
925 static const struct nl_policy odp_packet_policy[] = {
926 /* Always present. */
927 [ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
928 .min_len = ETH_HEADER_LEN },
929 [ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
931 /* ODP_PACKET_CMD_ACTION only. */
932 [ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
934 /* ODP_PACKET_CMD_SAMPLE only. */
935 [ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
936 [ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
939 struct odp_header *odp_header;
940 struct nlattr *a[ARRAY_SIZE(odp_packet_policy)];
941 struct nlmsghdr *nlmsg;
942 struct genlmsghdr *genl;
946 ofpbuf_use_const(&b, buf->data, buf->size);
948 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
949 genl = ofpbuf_try_pull(&b, sizeof *genl);
950 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
951 if (!nlmsg || !genl || !odp_header
952 || nlmsg->nlmsg_type != odp_packet_family
953 || !nl_policy_parse(&b, 0, odp_packet_policy, a,
954 ARRAY_SIZE(odp_packet_policy))) {
958 type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS
959 : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION
960 : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
966 memset(upcall, 0, sizeof *upcall);
968 upcall->packet = buf;
969 upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]);
970 upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]);
971 upcall->key = (void *) nl_attr_get(a[ODP_PACKET_ATTR_KEY]);
972 upcall->key_len = nl_attr_get_size(a[ODP_PACKET_ATTR_KEY]);
973 upcall->userdata = (a[ODP_PACKET_ATTR_USERDATA]
974 ? nl_attr_get_u64(a[ODP_PACKET_ATTR_USERDATA])
976 upcall->sample_pool = (a[ODP_PACKET_ATTR_SAMPLE_POOL]
977 ? nl_attr_get_u32(a[ODP_PACKET_ATTR_SAMPLE_POOL])
979 if (a[ODP_PACKET_ATTR_ACTIONS]) {
980 upcall->actions = (void *) nl_attr_get(a[ODP_PACKET_ATTR_ACTIONS]);
981 upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]);
984 *dp_ifindex = odp_header->dp_ifindex;
990 dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
992 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
997 if (!dpif->mc_sock) {
1001 for (i = 0; i < 50; i++) {
1004 error = nl_sock_recv(dpif->mc_sock, &buf, false);
1009 error = parse_odp_packet(buf, upcall, &dp_ifindex);
1011 && dp_ifindex == dpif->dp_ifindex
1012 && dpif->listen_mask & (1u << upcall->type)) {
1026 dpif_linux_recv_wait(struct dpif *dpif_)
1028 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1029 if (dpif->mc_sock) {
1030 nl_sock_wait(dpif->mc_sock, POLLIN);
1035 dpif_linux_recv_purge(struct dpif *dpif_)
1037 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1039 if (dpif->mc_sock) {
1040 nl_sock_drain(dpif->mc_sock);
1044 const struct dpif_class dpif_linux_class = {
1046 dpif_linux_enumerate,
1052 dpif_linux_get_stats,
1053 dpif_linux_get_drop_frags,
1054 dpif_linux_set_drop_frags,
1055 dpif_linux_port_add,
1056 dpif_linux_port_del,
1057 dpif_linux_port_query_by_number,
1058 dpif_linux_port_query_by_name,
1059 dpif_linux_get_max_ports,
1060 dpif_linux_port_dump_start,
1061 dpif_linux_port_dump_next,
1062 dpif_linux_port_dump_done,
1063 dpif_linux_port_poll,
1064 dpif_linux_port_poll_wait,
1065 dpif_linux_flow_get,
1066 dpif_linux_flow_put,
1067 dpif_linux_flow_del,
1068 dpif_linux_flow_flush,
1069 dpif_linux_flow_dump_start,
1070 dpif_linux_flow_dump_next,
1071 dpif_linux_flow_dump_done,
1073 dpif_linux_recv_get_mask,
1074 dpif_linux_recv_set_mask,
1075 dpif_linux_get_sflow_probability,
1076 dpif_linux_set_sflow_probability,
1077 dpif_linux_queue_to_priority,
1079 dpif_linux_recv_wait,
1080 dpif_linux_recv_purge,
1084 dpif_linux_init(void)
1086 static int error = -1;
1089 error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY,
1090 &odp_datapath_family);
1092 VLOG_ERR("Generic Netlink family '%s' does not exist. "
1093 "The Open vSwitch kernel module is probably not loaded.",
1094 ODP_DATAPATH_FAMILY);
1097 error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family);
1100 error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family);
1103 error = nl_lookup_genl_family(ODP_PACKET_FAMILY,
1104 &odp_packet_family);
1107 error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
1115 dpif_linux_is_internal_device(const char *name)
1117 struct dpif_linux_vport reply;
1121 error = dpif_linux_vport_get(name, &reply, &buf);
1124 } else if (error != ENODEV && error != ENOENT) {
1125 VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)",
1126 name, strerror(error));
1129 return reply.type == ODP_VPORT_TYPE_INTERNAL;
1133 dpif_linux_vport_send(int dp_ifindex, uint32_t port_no,
1134 const void *data, size_t size)
1136 struct ofpbuf actions, key, packet;
1137 struct odputil_keybuf keybuf;
1141 ofpbuf_use_const(&packet, data, size);
1142 flow_extract(&packet, htonll(0), 0, &flow);
1144 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1145 odp_flow_key_from_flow(&key, &flow);
1147 ofpbuf_use_stack(&actions, &action, sizeof action);
1148 nl_msg_put_u32(&actions, ODP_ACTION_ATTR_OUTPUT, port_no);
1150 return dpif_linux_execute__(dp_ifindex, key.data, key.size,
1151 actions.data, actions.size, &packet);
1155 dpif_linux_port_changed(const struct rtnetlink_link_change *change,
1158 struct dpif_linux *dpif = dpif_;
1161 if (change->master_ifindex == dpif->dp_ifindex
1162 && (change->nlmsg_type == RTM_NEWLINK
1163 || change->nlmsg_type == RTM_DELLINK))
1165 /* Our datapath changed, either adding a new port or deleting an
1167 sset_add(&dpif->changed_ports, change->ifname);
1170 dpif->change_error = true;
1174 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1175 * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
1176 * positive errno value.
1178 * 'vport' will contain pointers into 'buf', so the caller should not free
1179 * 'buf' while 'vport' is still in use. */
1181 dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport,
1182 const struct ofpbuf *buf)
1184 static const struct nl_policy odp_vport_policy[] = {
1185 [ODP_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
1186 [ODP_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
1187 [ODP_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1188 [ODP_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
1189 .min_len = sizeof(struct rtnl_link_stats64),
1190 .max_len = sizeof(struct rtnl_link_stats64),
1192 [ODP_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
1193 .min_len = ETH_ADDR_LEN,
1194 .max_len = ETH_ADDR_LEN,
1196 [ODP_VPORT_ATTR_MTU] = { .type = NL_A_U32, .optional = true },
1197 [ODP_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
1198 [ODP_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
1199 [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true },
1202 struct nlattr *a[ARRAY_SIZE(odp_vport_policy)];
1203 struct odp_header *odp_header;
1204 struct nlmsghdr *nlmsg;
1205 struct genlmsghdr *genl;
1208 dpif_linux_vport_init(vport);
1210 ofpbuf_use_const(&b, buf->data, buf->size);
1211 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1212 genl = ofpbuf_try_pull(&b, sizeof *genl);
1213 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1214 if (!nlmsg || !genl || !odp_header
1215 || nlmsg->nlmsg_type != odp_vport_family
1216 || !nl_policy_parse(&b, 0, odp_vport_policy, a,
1217 ARRAY_SIZE(odp_vport_policy))) {
1221 vport->cmd = genl->cmd;
1222 vport->dp_ifindex = odp_header->dp_ifindex;
1223 vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1224 vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1225 vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]);
1226 if (a[ODP_VPORT_ATTR_STATS]) {
1227 vport->stats = nl_attr_get(a[ODP_VPORT_ATTR_STATS]);
1229 if (a[ODP_VPORT_ATTR_ADDRESS]) {
1230 vport->address = nl_attr_get(a[ODP_VPORT_ATTR_ADDRESS]);
1232 if (a[ODP_VPORT_ATTR_MTU]) {
1233 vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]);
1235 vport->mtu = INT_MAX;
1237 if (a[ODP_VPORT_ATTR_OPTIONS]) {
1238 vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]);
1239 vport->options_len = nl_attr_get_size(a[ODP_VPORT_ATTR_OPTIONS]);
1241 if (a[ODP_VPORT_ATTR_IFINDEX]) {
1242 vport->ifindex = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFINDEX]);
1244 if (a[ODP_VPORT_ATTR_IFLINK]) {
1245 vport->iflink = nl_attr_get_u32(a[ODP_VPORT_ATTR_IFLINK]);
1250 /* Appends to 'buf' (which must initially be empty) a "struct odp_header"
1251 * followed by Netlink attributes corresponding to 'vport'. */
1253 dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport,
1256 struct odp_header *odp_header;
1258 nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
1261 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1262 odp_header->dp_ifindex = vport->dp_ifindex;
1264 if (vport->port_no != UINT32_MAX) {
1265 nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1268 if (vport->type != ODP_VPORT_TYPE_UNSPEC) {
1269 nl_msg_put_u32(buf, ODP_VPORT_ATTR_TYPE, vport->type);
1273 nl_msg_put_string(buf, ODP_VPORT_ATTR_NAME, vport->name);
1277 nl_msg_put_unspec(buf, ODP_VPORT_ATTR_STATS,
1278 vport->stats, sizeof *vport->stats);
1281 if (vport->address) {
1282 nl_msg_put_unspec(buf, ODP_VPORT_ATTR_ADDRESS,
1283 vport->address, ETH_ADDR_LEN);
1286 if (vport->mtu && vport->mtu != INT_MAX) {
1287 nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu);
1290 if (vport->options) {
1291 nl_msg_put_nested(buf, ODP_VPORT_ATTR_OPTIONS,
1292 vport->options, vport->options_len);
1295 if (vport->ifindex) {
1296 nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFINDEX, vport->ifindex);
1299 if (vport->iflink) {
1300 nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink);
1304 /* Clears 'vport' to "empty" values. */
1306 dpif_linux_vport_init(struct dpif_linux_vport *vport)
1308 memset(vport, 0, sizeof *vport);
1309 vport->port_no = UINT32_MAX;
1312 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1313 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1314 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1315 * result of the command is expected to be an odp_vport also, which is decoded
1316 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1317 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1319 dpif_linux_vport_transact(const struct dpif_linux_vport *request,
1320 struct dpif_linux_vport *reply,
1321 struct ofpbuf **bufp)
1323 struct ofpbuf *request_buf;
1326 assert((reply != NULL) == (bufp != NULL));
1328 error = dpif_linux_init();
1332 dpif_linux_vport_init(reply);
1337 request_buf = ofpbuf_new(1024);
1338 dpif_linux_vport_to_ofpbuf(request, request_buf);
1339 error = nl_sock_transact(genl_sock, request_buf, bufp);
1340 ofpbuf_delete(request_buf);
1344 error = dpif_linux_vport_from_ofpbuf(reply, *bufp);
1347 dpif_linux_vport_init(reply);
1348 ofpbuf_delete(*bufp);
1355 /* Obtains information about the kernel vport named 'name' and stores it into
1356 * '*reply' and '*bufp'. The caller must free '*bufp' when the reply is no
1357 * longer needed ('reply' will contain pointers into '*bufp'). */
1359 dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply,
1360 struct ofpbuf **bufp)
1362 struct dpif_linux_vport request;
1364 dpif_linux_vport_init(&request);
1365 request.cmd = ODP_VPORT_CMD_GET;
1366 request.name = name;
1368 return dpif_linux_vport_transact(&request, reply, bufp);
1371 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1372 * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
1373 * positive errno value.
1375 * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
1376 * while 'dp' is still in use. */
1378 dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
1380 static const struct nl_policy odp_datapath_policy[] = {
1381 [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
1382 [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
1383 .min_len = sizeof(struct odp_stats),
1384 .max_len = sizeof(struct odp_stats),
1386 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
1387 [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
1388 [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
1391 struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)];
1392 struct odp_header *odp_header;
1393 struct nlmsghdr *nlmsg;
1394 struct genlmsghdr *genl;
1397 dpif_linux_dp_init(dp);
1399 ofpbuf_use_const(&b, buf->data, buf->size);
1400 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1401 genl = ofpbuf_try_pull(&b, sizeof *genl);
1402 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1403 if (!nlmsg || !genl || !odp_header
1404 || nlmsg->nlmsg_type != odp_datapath_family
1405 || !nl_policy_parse(&b, 0, odp_datapath_policy, a,
1406 ARRAY_SIZE(odp_datapath_policy))) {
1410 dp->cmd = genl->cmd;
1411 dp->dp_ifindex = odp_header->dp_ifindex;
1412 dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
1413 if (a[ODP_DP_ATTR_STATS]) {
1414 /* Can't use structure assignment because Netlink doesn't ensure
1415 * sufficient alignment for 64-bit members. */
1416 memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]),
1419 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1420 dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1422 if (a[ODP_DP_ATTR_SAMPLING]) {
1423 dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]);
1426 if (a[ODP_DP_ATTR_MCGROUPS]) {
1427 static const struct nl_policy odp_mcgroup_policy[] = {
1428 [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
1429 [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
1430 [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
1433 struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)];
1435 if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy,
1436 mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) {
1440 if (mcgroups[ODP_PACKET_CMD_MISS]) {
1441 dp->mcgroups[DPIF_UC_MISS]
1442 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]);
1444 if (mcgroups[ODP_PACKET_CMD_ACTION]) {
1445 dp->mcgroups[DPIF_UC_ACTION]
1446 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]);
1448 if (mcgroups[ODP_PACKET_CMD_SAMPLE]) {
1449 dp->mcgroups[DPIF_UC_SAMPLE]
1450 = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]);
1457 /* Appends to 'buf' the Generic Netlink message described by 'dp'. */
1459 dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
1461 struct odp_header *odp_header;
1463 nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family,
1464 NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
1466 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1467 odp_header->dp_ifindex = dp->dp_ifindex;
1470 nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
1473 /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */
1475 if (dp->ipv4_frags) {
1476 nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
1480 nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
1484 /* Clears 'dp' to "empty" values. */
1486 dpif_linux_dp_init(struct dpif_linux_dp *dp)
1488 memset(dp, 0, sizeof *dp);
1492 dpif_linux_dp_dump_start(struct nl_dump *dump)
1494 struct dpif_linux_dp request;
1497 dpif_linux_dp_init(&request);
1498 request.cmd = ODP_DP_CMD_GET;
1500 buf = ofpbuf_new(1024);
1501 dpif_linux_dp_to_ofpbuf(&request, buf);
1502 nl_dump_start(dump, genl_sock, buf);
1506 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1507 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1508 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1509 * result of the command is expected to be of the same form, which is decoded
1510 * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
1511 * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
1513 dpif_linux_dp_transact(const struct dpif_linux_dp *request,
1514 struct dpif_linux_dp *reply, struct ofpbuf **bufp)
1516 struct ofpbuf *request_buf;
1519 assert((reply != NULL) == (bufp != NULL));
1521 request_buf = ofpbuf_new(1024);
1522 dpif_linux_dp_to_ofpbuf(request, request_buf);
1523 error = nl_sock_transact(genl_sock, request_buf, bufp);
1524 ofpbuf_delete(request_buf);
1528 error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
1531 dpif_linux_dp_init(reply);
1532 ofpbuf_delete(*bufp);
1539 /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
1540 * The caller must free '*bufp' when the reply is no longer needed ('reply'
1541 * will contain pointers into '*bufp'). */
1543 dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
1544 struct ofpbuf **bufp)
1546 struct dpif_linux *dpif = dpif_linux_cast(dpif_);
1547 struct dpif_linux_dp request;
1549 dpif_linux_dp_init(&request);
1550 request.cmd = ODP_DP_CMD_GET;
1551 request.dp_ifindex = dpif->dp_ifindex;
1553 return dpif_linux_dp_transact(&request, reply, bufp);
1556 /* Parses the contents of 'buf', which contains a "struct odp_header" followed
1557 * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
1558 * positive errno value.
1560 * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
1561 * while 'flow' is still in use. */
1563 dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
1564 const struct ofpbuf *buf)
1566 static const struct nl_policy odp_flow_policy[] = {
1567 [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
1568 [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
1569 [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
1570 .min_len = sizeof(struct odp_flow_stats),
1571 .max_len = sizeof(struct odp_flow_stats),
1573 [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
1574 [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
1575 /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */
1578 struct nlattr *a[ARRAY_SIZE(odp_flow_policy)];
1579 struct odp_header *odp_header;
1580 struct nlmsghdr *nlmsg;
1581 struct genlmsghdr *genl;
1584 dpif_linux_flow_init(flow);
1586 ofpbuf_use_const(&b, buf->data, buf->size);
1587 nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
1588 genl = ofpbuf_try_pull(&b, sizeof *genl);
1589 odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
1590 if (!nlmsg || !genl || !odp_header
1591 || nlmsg->nlmsg_type != odp_flow_family
1592 || !nl_policy_parse(&b, 0, odp_flow_policy, a,
1593 ARRAY_SIZE(odp_flow_policy))) {
1597 flow->nlmsg_flags = nlmsg->nlmsg_flags;
1598 flow->dp_ifindex = odp_header->dp_ifindex;
1599 flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]);
1600 flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]);
1601 if (a[ODP_FLOW_ATTR_ACTIONS]) {
1602 flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]);
1603 flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]);
1605 if (a[ODP_FLOW_ATTR_STATS]) {
1606 flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]);
1608 if (a[ODP_FLOW_ATTR_TCP_FLAGS]) {
1609 flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]);
1611 if (a[ODP_FLOW_ATTR_USED]) {
1612 flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]);
1617 /* Appends to 'buf' (which must initially be empty) a "struct odp_header"
1618 * followed by Netlink attributes corresponding to 'flow'. */
1620 dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
1623 struct odp_header *odp_header;
1625 nl_msg_put_genlmsghdr(buf, 0, odp_flow_family,
1626 NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
1629 odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
1630 odp_header->dp_ifindex = flow->dp_ifindex;
1632 if (flow->key_len) {
1633 nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len);
1636 if (flow->actions || flow->actions_len) {
1637 nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS,
1638 flow->actions, flow->actions_len);
1641 /* We never need to send these to the kernel. */
1642 assert(!flow->stats);
1643 assert(!flow->tcp_flags);
1644 assert(!flow->used);
1647 nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
1651 /* Clears 'flow' to "empty" values. */
1653 dpif_linux_flow_init(struct dpif_linux_flow *flow)
1655 memset(flow, 0, sizeof *flow);
1658 /* Executes 'request' in the kernel datapath. If the command fails, returns a
1659 * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
1660 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
1661 * result of the command is expected to be a flow also, which is decoded and
1662 * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
1663 * is no longer needed ('reply' will contain pointers into '*bufp'). */
1665 dpif_linux_flow_transact(const struct dpif_linux_flow *request,
1666 struct dpif_linux_flow *reply, struct ofpbuf **bufp)
1668 struct ofpbuf *request_buf;
1671 assert((reply != NULL) == (bufp != NULL));
1673 request_buf = ofpbuf_new(1024);
1674 dpif_linux_flow_to_ofpbuf(request, request_buf);
1675 error = nl_sock_transact(genl_sock, request_buf, bufp);
1676 ofpbuf_delete(request_buf);
1680 error = dpif_linux_flow_from_ofpbuf(reply, *bufp);
1683 dpif_linux_flow_init(reply);
1684 ofpbuf_delete(*bufp);
1692 dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
1693 struct dpif_flow_stats *stats)
1696 stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
1697 stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
1699 stats->n_packets = 0;
1702 stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
1703 stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;